Most refinement algorithms for single particle reconstruction in Cryo-EM require a low resolution initial model and an approximation of the poses in each particle image as a good initialization. Therefore, having a high quality ab initio model is of crucial importance. Often, this is computed using class averages, obtained by splitting the particle images into different classes corresponding to different orientations and averaging each cluster, a process which leads to denoised particle images, but also to loss of information. In this proof-of-concept work, we show how Markov Chain Monte-Carlo (MCMC) sampling methods can be used to obtain an initial model directly from the noisy particle images, bypassing the need for class averaging. In this talk, we describe useful sampling strategies of the volume and the poses, highlighting the benefits of this method and show results on both simulated and experimental data.
Back to Computational Microscopy